From 5360cb3d5608ab375de6cd8c0b408459f3fa953a Mon Sep 17 00:00:00 2001 From: Gregory Szorc Date: Mon, 22 May 2023 04:45:20 -0700 Subject: [PATCH] gh-101282: Apply BOLT optimizations to libpython for shared builds (#104709) Apply BOLT optimizations to libpython for shared builds. Most of the C code is in libpython so it is critical to apply BOLT there fully realize BOLT benefits. This change also reworks how BOLT instrumentation is applied. It effectively removes the readelf based logic added in gh-101525 and replaces it with a mechanism that saves a copy of the pre-bolt binary and restores that copy when necessary. This allows us to perform BOLT optimizations without having to manually delete the output binary to force a new bolt run. Also: - add a clean-bolt target for purging BOLT files and hook that up to the clean target - .gitignore BOLT related files Before and after this refactor, `make` will no-op after a previous run. Both versions should also share common make DAG deficiencies where targets fail to trigger as often as they need to or can trigger prematurely in certain scenarios. e.g. after this change you may need to `rm profile-bolt-stamp` to force a BOLT run because there aren't appropriate non-phony targets for BOLT's make target to depend on. To make it easier to iterate on custom BOLT settings, the flags to pass to instrumentation and application are now defined in configure and can be overridden by passing BOLT_INSTRUMENT_FLAGS and BOLT_APPLY_FLAGS. --- .gitignore | 5 + Doc/using/configure.rst | 7 + Makefile.pre.in | 67 ++++++-- ...-05-20-16-09-59.gh-issue-101282.FvRARb.rst | 4 + configure | 147 +++++------------- configure.ac | 55 ++++++- 6 files changed, 154 insertions(+), 131 deletions(-) create mode 100644 Misc/NEWS.d/next/Build/2023-05-20-16-09-59.gh-issue-101282.FvRARb.rst diff --git a/.gitignore b/.gitignore index d9c4a7972f0..ef7642b09bc 100644 --- a/.gitignore +++ b/.gitignore @@ -23,6 +23,10 @@ *.gc?? *.profclang? *.profraw +# Copies of binaries before BOLT optimizations. +*.prebolt +# BOLT profile data. +*.fdata *.dyn .gdb_history .purify @@ -124,6 +128,7 @@ Tools/unicode/data/ /platform /profile-clean-stamp /profile-run-stamp +/profile-bolt-stamp /Python/deepfreeze/*.c /pybuilddir.txt /pyconfig.h diff --git a/Doc/using/configure.rst b/Doc/using/configure.rst index ce858ab2c8d..fbe280d6413 100644 --- a/Doc/using/configure.rst +++ b/Doc/using/configure.rst @@ -314,6 +314,13 @@ also be used to improve performance. is dependent on a combination of the build environment + the other optimization configure args + the CPU architecture, and not all combinations are supported. + BOLT versions before LLVM 16 are known to crash BOLT under some scenarios. + Use of LLVM 16 or newer for BOLT optimization is strongly encouraged. + + The :envvar:`!BOLT_INSTRUMENT_FLAGS` and :envvar:`!BOLT_APPLY_FLAGS` + :program:`configure` variables can be defined to override the default set of + arguments for :program:`llvm-bolt` to instrument and apply BOLT data to + binaries, respectively. .. versionadded:: 3.12 diff --git a/Makefile.pre.in b/Makefile.pre.in index da3a8f6c13f..eb79c9c4ca1 100644 --- a/Makefile.pre.in +++ b/Makefile.pre.in @@ -672,21 +672,55 @@ profile-opt: profile-run-stamp -rm -f profile-clean-stamp $(MAKE) @DEF_MAKE_RULE@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_USE_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST)" -.PHONY: bolt-opt -bolt-opt: @PREBOLT_RULE@ - rm -f *.fdata - @if $(READELF) -p .note.bolt_info $(BUILDPYTHON) | grep BOLT > /dev/null; then\ - echo "skip: $(BUILDPYTHON) is already BOLTed."; \ - else \ - @LLVM_BOLT@ ./$(BUILDPYTHON) -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $(BUILDPYTHON).bolt) -o $(BUILDPYTHON).bolt_inst; \ - ./$(BUILDPYTHON).bolt_inst $(PROFILE_TASK) || true; \ - @MERGE_FDATA@ $(BUILDPYTHON).*.fdata > $(BUILDPYTHON).fdata; \ - @LLVM_BOLT@ ./$(BUILDPYTHON) -o $(BUILDPYTHON).bolt -data=$(BUILDPYTHON).fdata -update-debug-sections -reorder-blocks=ext-tsp -reorder-functions=hfsort+ -split-functions -icf=1 -inline-all -split-eh -reorder-functions-use-hot-size -peepholes=none -jump-tables=aggressive -inline-ap -indirect-call-promotion=all -dyno-stats -use-gnu-stack -frame-opt=hot; \ - rm -f *.fdata; \ - rm -f $(BUILDPYTHON).bolt_inst; \ - mv $(BUILDPYTHON).bolt $(BUILDPYTHON); \ - fi +# List of binaries that BOLT runs on. +BOLT_BINARIES := @BOLT_BINARIES@ +BOLT_INSTRUMENT_FLAGS := @BOLT_INSTRUMENT_FLAGS@ +BOLT_APPLY_FLAGS := @BOLT_APPLY_FLAGS@ + +.PHONY: clean-bolt +clean-bolt: + # Profile data. + rm -f *.fdata + # Pristine binaries before BOLT optimization. + rm -f *.prebolt + # BOLT instrumented binaries. + rm -f *.bolt_inst + +profile-bolt-stamp: $(BUILDPYTHON) + # Ensure a pristine, pre-BOLT copy of the binary and no profile data from last run. + for bin in $(BOLT_BINARIES); do \ + prebolt="$${bin}.prebolt"; \ + if [ -e "$${prebolt}" ]; then \ + echo "Restoring pre-BOLT binary $${prebolt}"; \ + mv "$${bin}.prebolt" "$${bin}"; \ + fi; \ + cp "$${bin}" "$${prebolt}"; \ + rm -f $${bin}.bolt.*.fdata $${bin}.fdata; \ + done + # Instrument each binary. + for bin in $(BOLT_BINARIES); do \ + @LLVM_BOLT@ "$${bin}" -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $${bin}.bolt) -o $${bin}.bolt_inst $(BOLT_INSTRUMENT_FLAGS); \ + mv "$${bin}.bolt_inst" "$${bin}"; \ + done + # Run instrumented binaries to collect data. + $(RUNSHARED) ./$(BUILDPYTHON) $(PROFILE_TASK) || true + # Merge all the data files together. + for bin in $(BOLT_BINARIES); do \ + @MERGE_FDATA@ $${bin}.*.fdata > "$${bin}.fdata"; \ + rm -f $${bin}.*.fdata; \ + done + # Run bolt against the merged data to produce an optimized binary. + for bin in $(BOLT_BINARIES); do \ + @LLVM_BOLT@ "$${bin}.prebolt" -o "$${bin}.bolt" -data="$${bin}.fdata" $(BOLT_APPLY_FLAGS); \ + mv "$${bin}.bolt" "$${bin}"; \ + done + touch $@ + +.PHONY: bolt-opt +bolt-opt: + $(MAKE) @PREBOLT_RULE@ + $(MAKE) profile-bolt-stamp # Compile and run with gcov .PHONY: coverage @@ -2623,10 +2657,11 @@ profile-removal: rm -f $(COVERAGE_INFO) rm -rf $(COVERAGE_REPORT) rm -f profile-run-stamp + rm -f profile-bolt-stamp .PHONY: clean -clean: clean-retain-profile - @if test @DEF_MAKE_ALL_RULE@ = profile-opt; then \ +clean: clean-retain-profile clean-bolt + @if test @DEF_MAKE_ALL_RULE@ = profile-opt -o @DEF_MAKE_ALL_RULE@ = bolt-opt; then \ rm -f profile-gen-stamp profile-clean-stamp; \ $(MAKE) profile-removal; \ fi diff --git a/Misc/NEWS.d/next/Build/2023-05-20-16-09-59.gh-issue-101282.FvRARb.rst b/Misc/NEWS.d/next/Build/2023-05-20-16-09-59.gh-issue-101282.FvRARb.rst new file mode 100644 index 00000000000..cc70d47c6c1 --- /dev/null +++ b/Misc/NEWS.d/next/Build/2023-05-20-16-09-59.gh-issue-101282.FvRARb.rst @@ -0,0 +1,4 @@ +BOLT optimization is now applied to the libpython shared library if building +a shared library. BOLT instrumentation and application settings can now be +influenced via the ``BOLT_INSTRUMENT_FLAGS`` and ``BOLT_APPLY_FLAGS`` +configure variables. diff --git a/configure b/configure index 7aad4fe89e3..2b863be108b 100755 --- a/configure +++ b/configure @@ -883,10 +883,11 @@ CFLAGS_NODIST BASECFLAGS CFLAGS_ALIASING OPT +BOLT_APPLY_FLAGS +BOLT_INSTRUMENT_FLAGS +BOLT_BINARIES MERGE_FDATA LLVM_BOLT -ac_ct_READELF -READELF PREBOLT_RULE LLVM_PROF_FOUND LLVM_PROFDATA @@ -1105,6 +1106,8 @@ CPPFLAGS CPP HOSTRUNNER PROFILE_TASK +BOLT_INSTRUMENT_FLAGS +BOLT_APPLY_FLAGS LIBUUID_CFLAGS LIBUUID_LIBS LIBFFI_CFLAGS @@ -1916,6 +1919,10 @@ Some influential environment variables: HOSTRUNNER Program to run CPython for the host platform PROFILE_TASK Python args for PGO generation task + BOLT_INSTRUMENT_FLAGS + Arguments to llvm-bolt when instrumenting binaries + BOLT_APPLY_FLAGS + Arguments to llvm-bolt when creating a BOLT optimized binary LIBUUID_CFLAGS C compiler flags for LIBUUID, overriding pkg-config LIBUUID_LIBS @@ -8106,112 +8113,6 @@ if test "$Py_BOLT" = 'true' ; then DEF_MAKE_ALL_RULE="bolt-opt" DEF_MAKE_RULE="build_all" - - if test -n "$ac_tool_prefix"; then - for ac_prog in readelf - do - # Extract the first word of "$ac_tool_prefix$ac_prog", so it can be a program name with args. -set dummy $ac_tool_prefix$ac_prog; ac_word=$2 -{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5 -$as_echo_n "checking for $ac_word... " >&6; } -if ${ac_cv_prog_READELF+:} false; then : - $as_echo_n "(cached) " >&6 -else - if test -n "$READELF"; then - ac_cv_prog_READELF="$READELF" # Let the user override the test. -else -as_save_IFS=$IFS; IFS=$PATH_SEPARATOR -for as_dir in $PATH -do - IFS=$as_save_IFS - test -z "$as_dir" && as_dir=. - for ac_exec_ext in '' $ac_executable_extensions; do - if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then - ac_cv_prog_READELF="$ac_tool_prefix$ac_prog" - $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5 - break 2 - fi -done - done -IFS=$as_save_IFS - -fi -fi -READELF=$ac_cv_prog_READELF -if test -n "$READELF"; then - { $as_echo "$as_me:${as_lineno-$LINENO}: result: $READELF" >&5 -$as_echo "$READELF" >&6; } -else - { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5 -$as_echo "no" >&6; } -fi - - - test -n "$READELF" && break - done -fi -if test -z "$READELF"; then - ac_ct_READELF=$READELF - for ac_prog in readelf -do - # Extract the first word of "$ac_prog", so it can be a program name with args. -set dummy $ac_prog; ac_word=$2 -{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5 -$as_echo_n "checking for $ac_word... " >&6; } -if ${ac_cv_prog_ac_ct_READELF+:} false; then : - $as_echo_n "(cached) " >&6 -else - if test -n "$ac_ct_READELF"; then - ac_cv_prog_ac_ct_READELF="$ac_ct_READELF" # Let the user override the test. -else -as_save_IFS=$IFS; IFS=$PATH_SEPARATOR -for as_dir in $PATH -do - IFS=$as_save_IFS - test -z "$as_dir" && as_dir=. - for ac_exec_ext in '' $ac_executable_extensions; do - if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then - ac_cv_prog_ac_ct_READELF="$ac_prog" - $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5 - break 2 - fi -done - done -IFS=$as_save_IFS - -fi -fi -ac_ct_READELF=$ac_cv_prog_ac_ct_READELF -if test -n "$ac_ct_READELF"; then - { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_ct_READELF" >&5 -$as_echo "$ac_ct_READELF" >&6; } -else - { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5 -$as_echo "no" >&6; } -fi - - - test -n "$ac_ct_READELF" && break -done - - if test "x$ac_ct_READELF" = x; then - READELF=""notfound"" - else - case $cross_compiling:$ac_tool_warned in -yes:) -{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: using cross tools not prefixed with host triplet" >&5 -$as_echo "$as_me: WARNING: using cross tools not prefixed with host triplet" >&2;} -ac_tool_warned=yes ;; -esac - READELF=$ac_ct_READELF - fi -fi - - if test "$READELF" == "notfound" - then - as_fn_error $? "readelf is required for a --enable-bolt build but could not be found." "$LINENO" 5 - fi - # -fno-reorder-blocks-and-partition is required for bolt to work. # Possibly GCC only. { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether C compiler accepts -fno-reorder-blocks-and-partition" >&5 @@ -8474,6 +8375,36 @@ $as_echo "\"Found merge-fdata\"" >&6; } fi fi + +BOLT_BINARIES='$(BUILDPYTHON)' +if test "x$enable_shared" = xyes; then : + + BOLT_BINARIES="${BOLT_BINARIES} \$(INSTSONAME)" + +fi + + +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking BOLT_INSTRUMENT_FLAGS" >&5 +$as_echo_n "checking BOLT_INSTRUMENT_FLAGS... " >&6; } +if test -z "${BOLT_INSTRUMENT_FLAGS}" +then + BOLT_INSTRUMENT_FLAGS= +fi +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $BOLT_INSTRUMENT_FLAGS" >&5 +$as_echo "$BOLT_INSTRUMENT_FLAGS" >&6; } + + +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking BOLT_APPLY_FLAGS" >&5 +$as_echo_n "checking BOLT_APPLY_FLAGS... " >&6; } +if test -z "${BOLT_APPLY_FLAGS}" +then + BOLT_APPLY_FLAGS=-update-debug-sections -reorder-blocks=ext-tsp -reorder-functions=hfsort+ -split-functions -icf=1 -inline-all -split-eh -reorder-functions-use-hot-size -peepholes=none -jump-tables=aggressive -inline-ap -indirect-call-promotion=all -dyno-stats -use-gnu-stack -frame-opt=hot + + +fi +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $BOLT_APPLY_FLAGS" >&5 +$as_echo "$BOLT_APPLY_FLAGS" >&6; } + # XXX Shouldn't the code above that fiddles with BASECFLAGS and OPT be # merged with this chunk of code? diff --git a/configure.ac b/configure.ac index 115998e0753..786d3414eb0 100644 --- a/configure.ac +++ b/configure.ac @@ -2028,13 +2028,6 @@ if test "$Py_BOLT" = 'true' ; then DEF_MAKE_ALL_RULE="bolt-opt" DEF_MAKE_RULE="build_all" - AC_SUBST(READELF) - AC_CHECK_TOOLS(READELF, [readelf], "notfound") - if test "$READELF" == "notfound" - then - AC_MSG_ERROR([readelf is required for a --enable-bolt build but could not be found.]) - fi - # -fno-reorder-blocks-and-partition is required for bolt to work. # Possibly GCC only. AX_CHECK_COMPILE_FLAG([-fno-reorder-blocks-and-partition],[ @@ -2067,6 +2060,54 @@ if test "$Py_BOLT" = 'true' ; then fi fi +dnl Enable BOLT of libpython if built. +AC_SUBST(BOLT_BINARIES) +BOLT_BINARIES='$(BUILDPYTHON)' +AS_VAR_IF([enable_shared], [yes], [ + BOLT_BINARIES="${BOLT_BINARIES} \$(INSTSONAME)" +]) + +AC_ARG_VAR( + [BOLT_INSTRUMENT_FLAGS], + [Arguments to llvm-bolt when instrumenting binaries] +) +AC_MSG_CHECKING([BOLT_INSTRUMENT_FLAGS]) +if test -z "${BOLT_INSTRUMENT_FLAGS}" +then + BOLT_INSTRUMENT_FLAGS= +fi +AC_MSG_RESULT([$BOLT_INSTRUMENT_FLAGS]) + +AC_ARG_VAR( + [BOLT_APPLY_FLAGS], + [Arguments to llvm-bolt when creating a BOLT optimized binary] +) +AC_MSG_CHECKING([BOLT_APPLY_FLAGS]) +if test -z "${BOLT_APPLY_FLAGS}" +then + AS_VAR_SET( + [BOLT_APPLY_FLAGS], + [m4_join([ ], + [-update-debug-sections], + [-reorder-blocks=ext-tsp], + [-reorder-functions=hfsort+], + [-split-functions], + [-icf=1], + [-inline-all], + [-split-eh], + [-reorder-functions-use-hot-size], + [-peepholes=none], + [-jump-tables=aggressive], + [-inline-ap], + [-indirect-call-promotion=all], + [-dyno-stats], + [-use-gnu-stack], + [-frame-opt=hot] + )] + ) +fi +AC_MSG_RESULT([$BOLT_APPLY_FLAGS]) + # XXX Shouldn't the code above that fiddles with BASECFLAGS and OPT be # merged with this chunk of code?